Are Enterprise Servers Ready for the Demands of Agentic AI?

Key Highlights

• HPE has announced four new ProLiant Gen13 servers powered by 6th Gen AMD EPYC processors.
• The release targets compute-intensive workloads such as agentic AI, inference, and complex data modelling.
• New integrated security features include automatic self-encrypting drives and multi-party authorization from first boot.
• High-density cooling and PCIe Gen6 connectivity aim to support the scaling requirements of next-generation enterprise applications.
• The announcement addresses the growing need for trusted infrastructure to mitigate operational risk at scale.

The News

HPE has expanded its enterprise server portfolio with the introduction of four new ProLiant Gen13 systems featuring 6th Generation AMD EPYC processors. These new servers focus heavily on compute density, memory bandwidth, and integrated security capabilities to manage the rise of agentic AI and high-performance inference. The new lineup includes the DL585a, DL525, XD245, and XD285 servers, each addressing distinct enterprise environments and power management requirements. Find out more by clicking here to read the press release: https://www.hpe.com/us/en/newsroom/press-release/2026/10/hpe-accelerates-enterprise-ai-innovation-with-next-generation-amd-cpu-powered-servers.html

Analyst Take

The enterprise computing market is undergoing a significant architectural transition. Organisations are moving beyond the initial experimentation phase of generative models and are now looking to deploy autonomous, agentic AI. This changes the hardware equation. Traditional servers were built for general-purpose workloads, but autonomous agents require constant data processing and massive memory bandwidth. The gap between software ambitions and data center realities is widening. Power remains a huge bottleneck. Cooling limits are constraining deployments across the industry.

Security presents an equally daunting challenge. A recent McKinsey study highlighted that nearly two-thirds of organisations view risk and security as the top barriers to scaling agentic AI. When AI systems transition from generating text to executing autonomous actions, the potential blast radius of a security breach expands exponentially. Trust becomes a requirement.

HPE is responding to these challenges by baking security and thermal management directly into the hardware layer. We see the integration of AMD EPYC processors as a strategic move to address both compute density and energy efficiency. AMD's core counts let IT departments consolidate workloads, reducing the physical footprint needed for intensive inference tasks. HyperFRAME Lens data validates this infrastructure friction: 60.7% of organizations cite physical infrastructure as a "very significant" challenge to scaling AI, a key reason only 23% of enterprise AI/ML projects in the past year reached production and met their intended ROI.

What Was Announced

HPE revealed four ProLiant Gen13 server models powered by 6th Generation AMD EPYC processors. The HPE ProLiant DL585a Gen13 Server is a high-density 10U platform architected to accommodate up to eight double-wide PCIe GPUs and dual processors. It features PCIe Gen6 connectivity and aims to deliver high token throughput for tasks like retrieval-augmented generation fine-tuning. The HPE ProLiant DL525 Gen13 Server is a compact 1U single-socket system designed to support up to 256 cores. It provides 16-channel memory with support for high-speed DDR5 and MRDIMM. This unit is architected to handle workloads requiring extreme data throughput, such as fraud detection and market-data processing.

For high-performance computing, the HPE ProLiant XD245 Server uses a one-Open Rack Unit footprint with a four-node direct liquid-cooled design. It is designed to run up to eight AMD processors for high compute density. The HPE ProLiant XD285 Server provides a two-Open Rack Unit two-node, air-cooled design to cater to traditional data center environments. HPE also announced security updates through HPE iLO 8. These features include auto-SED capabilities, designed to protect data at rest from the first boot without external key managers. The servers introduce multi-party authorization for high-impact actions and post-quantum cryptography readiness extending beyond the firmware layer.

Memory bandwidth frequently chokes AI application performance before the processor reaches its limit. The inclusion of MRDIMM support aims to deliver higher data transfer rates to keep the massive core counts fed. The transition to PCIe Gen6 connectivity is another forward-looking architectural decision. It doubles the bandwidth of the previous generation. This is crucial for GPU-accelerated workloads, ensuring that data moves freely between the storage, memory, and accelerators.

One of the most pressing issues we observe in client environments is power management. The XD245 server integrates direct liquid cooling to handle the thermal output of the latest AMD chips. Air cooling is rapidly reaching its physical limits for high-density compute. The XD285 offers an air-cooled alternative, which is highly practical for organisations unable to retrofit their facilities for liquid cooling in the near term. This provides necessary flexibility.

Security must be built in. It is often treated as an afterthought in hardware design, patched with software overlays. HPE is shifting this paradigm with the iLO 8 management processor. The automatic self-encrypting drive functionality reduces the administrative burden on IT teams. Multi-party authorization brings a stringent compliance model directly to server management. It acts as a dual-control mechanism for actions like secure erase. This mitigates the insider threat vector, which is a growing concern as agentic AI systems gain access to sensitive corporate data.

The integration with HPE Compute Ops Management adds a layer of operational efficiency. The software is designed to provide real-time power capacity insights and simplify lifecycle operations. Organisations need to manage fleets of these complex servers without linearly scaling their administrative headcount. We see this combination of hardware security and centralized software management as a strong value proposition for enterprises operating distributed IT environments.

HPE is positioning the Gen13 platform as a foundational element for hybrid infrastructure. The firm recognises that while public cloud consumption remains strong, a significant portion of AI inference will happen on-premises due to data sovereignty and latency requirements. The combination of HPE hardware and AMD silicon is architected to give enterprises a viable alternative to hyperscale cloud providers for specific high-intensity workloads. The focus on Open Compute Project standards for the XD series also indicates a nod to hyperscale design principles brought directly into the enterprise data center.

Looking Ahead

Based on what we are observing, the central theme moving forward is the maturation of AI infrastructure from experimental sandboxes to secure, production-ready enterprise deployments. The key trend that we will be looking out for is how traditional server vendors integrate high-density cooling and advanced cryptography to support autonomous agents. Our perspective is that power and cooling will dictate the pace of AI adoption just as much as processor speed.

The announcement from HPE and AMD demonstrates a clear strategy to capture the enterprise inference market. They are competing directly with players like Dell Technologies and Lenovo, who are also racing to provide dense, AI-optimized hardware. The tectonic shift here is the move toward liquid cooling as a standard requirement rather than an exotic edge case.

Going forward we are going to be closely monitoring how the company performs on executing its supply chain and delivery promises for these liquid-cooled architectures. Scaling the manufacturing and deployment of systems like the XD245 will be complex. Furthermore, we will be tracking how the transition to PCIe Gen6 and MRDIMM technologies translates to measurable business value for early adopters. The hardware foundation is solid. Success depends on execution. The ultimate test will be how easily enterprise software ecosystems can leverage these new hardware capabilities to run agentic AI reliably. We anticipate that enterprise buyers will increasingly demand certified reference architectures that bundle these servers with specific AI software stacks. Navigating this transition requires more than just powerful silicon; it requires deep ecosystem partnerships.